These technologies are already working together to accelerate the discovery of new antimicrobial medicines. One subset of next-gen AI, dubbed generative models, produces hypotheses about the final molecule needed for a specific new drug. These AI models don’t just search for known molecules with relevant properties, such as the ability to bind to and neutralize a virus or a bacterium, they are powerful enough to learn features of the underlying data and can suggest new molecules that have not yet been synthesized. This design, as opposed to searching capability, is particularly transformative because the number of possible suitable molecules is greater than the number of atoms in the universe, prohibitively large for search tasks.
Generative AI can navigate this vast chemical space to discover the right molecule faster than any human using conventional methods. AI modeling already supports research that could help patients with Parkinson’s disease, diabetes and chronic pain. For example, antimicrobial peptides (AMPs), for example, small protein-like compounds, is one solution that is the subject of intensive study. These molecules hold great promise as next-generation antibiotics because they are inherently less susceptible to resistance and are produced naturally as part of the innate immune system of living organisms.
In recent studies published in Nature Biomedical Engineering, 2021the AI-assisted search for new, effective, non-toxic peptides produced 20 promising novel candidates in just 48 days, a striking reduction compared to the conventional development times for new compounds.
Among these were two novel candidates used against Klebsiella pneumoniae, a bacterium frequently found in hospitals that causes pneumonia and bloodstream infections and has become increasingly resistant to conventional classes of antibiotics. Obtaining such a result with conventional research methods would take years.
AMPs already in commercial use
Collaborative work between IBM, Unilever, and STFC, which hosts one of IBM Research’s Discovery Accelerators at the Hartree Center in the UK, has recently helped researchers better understand AMPs. Unilever has already used that new knowledge to create consumer products that boost the effects of these natural-defense peptides.
And, in this Biophysical Journal paper, researchers demonstrated how small-molecule additives (organic compounds with low molecular weights) are able to make AMPs much more potent and efficient. Using advanced simulation methods, IBM researchers, in combination with experimental studies from Unilever, also identified new molecular mechanisms that could be responsible for this enhanced potency. This is a first-of-its-kind proof of principle that scientists will take forward in ongoing collaborations.
Boosting material discovery with AI Generative models and advanced computer simulations is part of a much larger strategy at IBM Research, dubbed Accelerated Discovery, where we use emerging computing technologies to boost the scientific method and its application to discovery. The aim is to greatly speed up the rate of discovery of new materials and drugs, whether it is in preparation for the next global crisis or to rapidly address the current and the inevitable future ones.
This is just one element of the loop comprising the revised scientific method, a cutting-edge transformation of the traditional linear approach to material discovery. Broadly, AI learns about the desired properties of a new material. Next, another type of AI, IBM’s Deep Search, combs through the existing knowledge on the manufacturing of this specific material, meaning all the previous research tucked away in patents and papers.
Generative models have the potential to create a new molecule
Following this, the generative models create a possible new molecule based on the existing data. Once done, we use a high-performance computer to simulate this new candidate molecule and the reactions it should have with its neighbors to make sure it performs as expected. In the future, a quantum computer could improve these molecular simulations even further.
The final step is AI-driven lab testing to experimentally validate the predictions and develop actual molecules. At IBM, we do this with a tool called RoboRXN, a small, fridge-sized chemistry lab that combines AI, cloud computing and robots to help researchers create new molecules anywhere at anytime. The combination of these approaches is well suited to tackle general ‘inverse design’ problems. Here, the task is to find or create for the first time a material with a desired property or function, as opposed to computing or measuring the properties of large numbers of candidates.
Proof that AI can go beyond the limits of classical computing
The antibiotic crisis is a particularly urgent example of a global inverse design challenge in need of a true paradigm shift towards the way we discover materials. The rapid progress in quantum computing and the development of quantum machine-learning techniques is now creating realistic prospects of extending the reach of artificial intelligence beyond the limitations of classical computing. Early examples show promise for quantum advantages in model training speed, classification tasks and prediction accuracy.
Overall, combining the most powerful emerging AI techniques (possibly with quantum acceleration) to learn features linked to antimicrobial activity with physical modeling at the molecular scale to reveal the modes of action is, arguably, the most promising route to creating these essential compounds faster than ever before.
The article originally appeared in the World Economic Forum.